During its inaugural developer conference in London, Anthropic announced upgrades to its Claude Managed Agents, introducing self-hosted sandboxes and Model Context Protocol (MCP) tunnels. This event signifies a key moment for the company as it expands its presence in Europe, catering to a growing developer community.
Self-Hosted Sandboxes: Enhanced Control for Enterprises
The self-hosted sandbox feature, now in public beta on the Claude Platform, enables organizations to run tools within environments they manage. This option is particularly attractive for enterprises aiming to meet strict security protocols while utilizing AI capabilities. By overseeing the sandbox environment, businesses can keep sensitive data within their internal networks or with trusted managed providers.
Anthropic has teamed up with several providers, including Cloudflare, Daytona, Modal, and Vercel, each offering unique computing and isolation methods. For example, Cloudflare uses microVMs for scalable sandbox management, while Daytona allows for stateful sandboxes that can be paused and resumed, making sure operational continuity. Modal specializes in AI workloads, providing container-based sandboxes with access to key CPU and GPU resources. Vercel, on the other hand, combines quick startup times with virtual machine security.
Rogo, an AI platform in institutional finance, showcases the practical use of these features. Strib Walker, their product head, highlighted how Claude Managed Agents simplifies the agent loop while Vercel's sandbox infrastructure enables customized workload configurations.
MCP Tunnels: Secured Internal Networking
In addition to sandbox environments, MCP tunnels mark another important advancement. These tunnels allow access to Model Context Protocol servers within a private network, protecting sensitive information from public exposure. By creating a lightweight gateway, organizations can establish a secure, single outbound connection that maintains end-to-end encryption, eliminating the need for changes to inbound firewalls or public endpoints.
Anthropic has made the management of these tunnels accessible through organization administrator controls in the Claude Console. This integration allows enterprises to maintain their existing workflows without requiring significant changes to their current Managed Agents setups.
Implications for the AI Sector
The launch of these features follows previous enhancements to the Claude Platform, including the initial rollout of Managed Agents and the introduction of built-in memory capabilities. As businesses increasingly depend on AI for various applications, these developments reflect a solid response to the market's demand for greater control and security in AI deployments.
Anthropic's approach aligns with a broader trend in the AI sector, where companies are emphasizing user autonomy and data security. With rising regulatory scrutiny surrounding data privacy, features like self-hosted sandboxes and MCP tunnels equip organizations with the necessary tools to comply with regulations while effectively utilizing AI technologies.
Looking Ahead
As Anthropic continues to introduce features that bolster security and operational control, the space within the AI sector is likely to evolve. Companies looking to adopt AI solutions will increasingly gravitate toward platforms that emphasize both functionality and security. The advancements in the Claude Platform may set a new benchmark for how AI services are delivered and managed, particularly for enterprises handling sensitive or proprietary information.
With ongoing innovations and a focus on developer needs, Anthropic seems well-positioned to capture a significant share of the enterprise AI market, creating an environment where businesses can innovate without sacrificing security.
